AGI

[Cadavre Exquis]

https://twitter.com/_akhaliq/status/1635690030684418059

https://openai.com/research/gpt-4

https://cdn.openai.com/papers/gpt-4.pdf

Saludos.

[javi2]

Para quien le interese, un curso introductorio al Deep Learning del MIT (En inglés)
http://introtodeeplearning.com/

Cada viernes cuelgan una nueva sesión. Desde el 10 de marzo hasta el 12 de mayo.

An efficient and high-intensity bootcamp designed to teach you the fundamentals of deep learning as quickly as possible!

[muyuu]

https://blog.mozilla.org/en/mozilla/mozilla-launches-responsible-ai-challenge/

MARCH 14, 2023
author
IMO UDOM
The last few months it has become clear that AI is no longer our future, but our present. Some of the most exciting ideas for the future of both the internet and the world involve AI solutions. This didn’t happen overnight, decades of work have gone into this moment. Mozilla has been working to make sure that the future of AI benefits humanity in the right ways by investing in the creation of trustworthy AI.

We want entrepreneurs and builders to join us in creating a future where AI is developed through this responsible lens. That’s why we are relaunching our Mozilla Builders program with the Responsible AI Challenge.

At Mozilla, we believe in AI: in its power, its commercial opportunity, and its potential to solve the world’s most challenging problems. But now is the moment to make sure that it is developed responsibly to serve society.

If you want to build (or are already building) AI solutions that are ambitious but also ethical and holistic, the Mozilla Builder’s Responsible AI Challenge is for you. We will be inviting the top nominees to join a gathering of the brightest technologists, business leaders and ethicists working on trustworthy AI to help get your ideas off the ground. Participants will also have access to mentorship from some of the best minds in the industry, the ability to meet key contributors in this community, and an opportunity to win some funding for their project.

Mozilla will be investing $50,000 into the top applications and projects, with a grand prize of $25,000 for the first place winner.

For more information, please visit here. Applications open up March 30, 2023.

[Cadavre Exquis]

Microsoft Lays Off Key AI Ethics Team, Report Says
Posted by BeauHD on Wednesday March 15, 2023 @06:00AM from the questionable-decisions dept.

According to Platformer, Microsoft's recent layoffs included its entire ethics and society team within the artificial intelligence organization. "The move leaves Microsoft without a dedicated team to ensure its AI principles are closely tied to product design at a time when the company is leading the charge to make AI tools available to the mainstream, current and former employees said." From the report:

Microsoft still maintains an active Office of Responsible AI, which is tasked with creating rules and principles to govern the company's AI initiatives. The company says its overall investment in responsibility work is increasing despite the recent layoffs.

But employees said the ethics and society team played a critical role in ensuring that the company's responsible AI principles are actually reflected in the design of the products that ship. "People would look at the principles coming out of the office of responsible AI and say, 'I don't know how this applies,'" one former employee says. "Our job was to show them and to create rules in areas where there were none."

In recent years, the team designed a role-playing game called Judgment Call that helped designers envision potential harms that could result from AI and discuss them during product development. It was part of a larger "responsible innovation toolkit" that the team posted publicly. More recently, the team has been working to identify risks posed by Microsoft's adoption of OpenAI's technology throughout its suite of products. The ethics and society team was at its largest in 2020, when it had roughly 30 employees including engineers, designers, and philosophers. In October, the team was cut to roughly seven people as part of a reorganization.

"Microsoft is committed to developing AI products and experiences safely and responsibly, and does so by investing in people, processes, and partnerships that prioritize this," the company said in a statement. "Over the past six years we have increased the number of people across our product teams and within the Office of Responsible AI who, along with all of us at Microsoft, are accountable for ensuring we put our AI principles into practice. [...] We appreciate the trailblazing work the ethics and society team did to help us on our ongoing responsible AI journey."

Saludos.

[muyuu]

Lo he puesto entero en el hilo del fin del trabajo, que creo que es marginalmente más apropiado. No lo repito aquí porque es muy largo, pero podéis ir a la primera cita, o al enlace mientras sigue online, para verlo todo.

5 partes en la versión completa, aquí solamente pondré un pequeño resumen para no spamear el foro

https://writings.stephenwolfram.com/2023/03/will-ais-take-all-our-jobs-and-end-human-history-or-not-well-its-complicated/

Will AIs Take All Our Jobs and End Human History—or Not? Well, It’s Complicated…

March 15, 2023

partes:

The Shock of ChatGPT

[...]

Intuition from the Computational Universe

[...]

“AIs Will Never Be Able to Do That”

[...]

Will There Be Anything Left for the Humans to Do?

[...]

Generalized Economics and the Concept of Progress

[...]

How Can We Tell the AIs What to Do?

[...]

A World Run by AIs

[...]

Governance in an AI World

[...]

So Will It End Badly?

[...]

Preparing for an AI World

[...]

Afterword: Looking at Some Actual Data

[...]

So here are pie charts of different job categories at 50-year intervals:


And, yes, in 1850 the US was firmly an agricultural economy, with just over half of all jobs being in agriculture. But as agriculture got more efficient—with the introduction of machinery, irrigation, better seeds, fertilizers, etc.—the fraction dropped dramatically, to just a few percent today.

[...]

Beyond the overall distribution of job categories, we can also look at trends in individual categories over time—with each one in a sense providing a certain window onto history:



One can definitely see cases where the number of jobs decreases as a result of automation. And this happens not only in areas like agriculture and mining, but also for example in finance (fewer clerks and bank tellers), as well as in sales and retail (online shopping). Sometimes—as in the case of manufacturing—there’s a decrease of jobs partly because of automation, and partly because the jobs move out of the US (mainly to countries with lower labor costs).

[...]

So what can we conclude? The data seems quite well aligned with what we discussed in more general terms above. Well-developed areas get automated and need to employ fewer people. But technology also opens up new areas, which employ additional people. And—as we might expect from computational irreducibility—things generally get progressively more complicated, with additional knowledge and organizational structure opening up more “frontiers” where people are needed. But even though there are sometimes “sudden inventions”, it still always seems to take decades (or effectively a generation) for there to be any dramatic change in the number of jobs. (The few sharp changes visible in the plots seem mostly to be associated with specific economic events, and—often related—changes in government policies.)

But in addition to the different jobs that get done, there’s also the question of how individual people spend their time each day. And—while it certainly doesn’t live up to my own (rather extreme) level of personal analytics—there’s a certain amount of data on this that’s been collected over the years (by getting time diaries from randomly sampled people) in the American Heritage Time Use Study. So here, for example, are plots based on this survey for how the amount of time spent on different broad activities has varied over the decades (the main line shows the mean—in hours—for each activity; the shaded areas indicate successive deciles):



And, yes, people are spending more time on “media & computing”, some mixture of watching TV, playing videogames, etc. Housework, at least for women, takes less time, presumably mostly as a result of automation (appliances, etc.). (“Leisure” is basically “hanging out” as well as hobbies and social, cultural, sporting events, etc.; “Civic” includes volunteer, religious, etc. activities.)

[Cadavre Exquis]

Microsoft | The Future of Work With AI - Microsoft March 2023 Event

Saludos.

[Cadavre Exquis]

Google Workspace | A new era for AI and Google Workspace

Microsoft 365 | Introducing Microsoft 365 Copilot | Your Copilot for Work

Saludos.

[Cadavre Exquis]

AI Explained | The Model That Changes Everything: Alpaca Breakthrough (ft. Apple's LLM, BritGPT, Ernie and AlexaTM)

8 years of cost reduction in 5 weeks: how Stanford's Alpaca model changes everything, including the economics of OpenAI and GPT 4. The breakthrough, using self-instruct, has big implications for Apple's secret large language model, Baidu's ErnieBot, Amazon's attempts and even governmental efforts, like the newly announced BritGPT.

I will go through how Stanford put the model together, why it costs so little, and demonstrate in action versus ChatGPT and GPT 4. And what are the implications of short-circuiting human annotation like this? With analysis of a tweet by Eliezer Yudkowsky, I delve into the workings of the model and the questions it rises.

Saludos.

[Cadavre Exquis]

https://twitter.com/iScienceLuvr/status/1636479850214232064

Saludos.

[Cadavre Exquis]

https://twitter.com/michalkosinski/status/1636683810631974912

Saludos.

P.D. Como explica su autor más adelante en el mismo hilo, la cuestión no es si GPT-4 es consciente o no, sino el hecho de que esté dispuesto a jugar a un juego en el que, en la práctica, toma el control del ordenador del usuario.

https://twitter.com/michalkosinski/status/1636799287337295872

[muyuu]

https://www.justus.pw/posts/2023-03-19-this-time-its-different.html

This Time It's Different
Published:
March 19, 2023
With the recent hype around ChatGPT, GPT-4, and large language models in general, communities like Hacker News are breaking out into a this time it’s different doom and gloom.

While no one can predict the future, we can at least think about some possibilities for what can happen and maybe think about what is less likely to happen.

To summarize, I do not see us approaching the Singularity soon. Still, the recent progress, especially around GPT-4, suggests that analogous to the word processor, spellchecking, code reuse, and other time-saving and accuracy-improving measures will make a significant chunk of our tedious day-to-day work superfluous and allow us to focus on the more essential things (vs. the urgent, but not important). Who is us in this context? We are knowledge workers. We use our knowledge of this world and intellect to shape information and communicate it to others. We devise new contraptions and schemes and strive to improve or exploit something.

There were many moments in history where at least some people must have thought: This is it. There will be no more work left for us to perform. Let’s look at some of these moments and see how they turned out.

A great example many like to bring up is the fate of the farrier. Before the automobile, horses were, for many, the primary method of transportation, both for private and commercial purposes. Farriers were indispensable in allowing horses to get around where they needed to be, caring for their hooves, and making horseshoes. While the automobile industry grew, the horse-based transportation industry did not. While horses are still used for many purposes to this day, though decidedly less for the economy in its entirety, the profession of the farrier still needs to reap all the rewards of global economic growth. I don’t think it’s fair to say, as is often assessed, that the automobile brought about the demise of the farrier. On the other hand, they did not end up as winners either.

What about the printing press or any other of our advances in writing, such as the typewriter, spellcheckers, emails, grammar checkers, and so on? While the work of a classical secretary became unnecessary in many ways, the fact that the individual increase in productivity meant an overall growth of the economy meant, in turn, that those who would have lost their current role or even employment to technological advancement would find new occupations within this growth.

Similar things can be said about the loom, the steam engine, and other technologies. Looking at it from a long-term perspective, we see that they allowed our economic output to grow steadily. It brings us to a point where more smartphones and other gadgets than we ever need are being produced and then chucked onto a landfill one or two years later.

One of our guiding economic and social principles is that people are expected to work unless they have a convincing excuse not to. This produces phenomenons such as the bullshit job. In general, this environment really motivates people to continuously find new reasons to keep someone on a salary, be it to press the same button for eight hours a day or generally lounge around and look busy. Other than salaried employment, there is often no chance of receiving social insurance once old age, disability, or another circumstance prevents someone from working.

What about the phenomenon of outsourcing or off-shoring? Take the allegedly important voter category of the coal miner. Some politicians like advertising to coal miners and claim they will bring back thousands of coal mining jobs if elected. The reality is, especially nowadays, that coal is not needed where it used to be, and laborers in other countries work in readily exploitable conditions, such that coal can be procured much cheaper elsewhere. If globalization was a technology, it did not make the need for coal unnecessary. Businesses can now kick down on people on a global scale instead of kicking down on your local population.

Knowledge workers for most of the 20th century and for a few good years in the 21st century saw the advent of productivity technologies as force multipliers. The search engine did not make the legal profession unnecessary. They could now research 3 cases in the time it took them to research one case. While before, I would be busy checking out several books about programming from the library, I can now just kagi for whatever I need, and it’s right there.

Enter large language models. I don’t see them as anything other than force multipliers. The dull routine work of getting information is getting an upgrade once again. LLMs seem to have other uses as well. They follow instructions very well. I can ask ChatGPT to improve my marketing copy. If you give large language models tools, such as LangChain, they can become your junior developer. At least, this is my idea for a future project. Short-term, I don’t see LLMs replacing employees, but it certainly will allow us to idly stare at screens even more, pretending to work and hoping that looking at the wall clock will make time pass faster.

That is not to say that some future manifestation of artificial intelligence won’t partially or entirely alleviate the need to perform labor. At this point in time, I don’t see us being on such a dramatic growth trajectory. On the other hand, you really don’t want to be on the wrong side of history here and categorically deny that this version of the future is entirely impossible.

The nature and exact value of knowledge work have always been dubious. After all, I might get a great idea while strolling to the grocery store. Will that count as hours worked? Paying someone for their time is easier to justify when you see someone pulling a lever. That suggests one of the reasons the subjugators of our internet economy are again asking their people to come back to the office is so that they can satisfactorily measure each worker’s butt-on-seat time.

Then, when LLMs take over every mundane task we work on and sit at our desks in the office with absolutely nothing to do, we must devise new ways to look busy. OpenAI may sell a crank that you have to turn, as otherwise, the token generation speed of ChatGPT will grind to a halt. Sooner or later, we must ask ourselves what the point of most knowledge work is. It seemed dubious from the start that someone sitting in front of a laptop answering emails is paying their rent doing that, but being part of the chain of labor exploitation would guarantee some form of compensation for intense screen staring and computer touching.

Before you think I sound defeatist, ChatGPT-3.5 absolved me:

While there are some elements of skepticism and critical reflection in your blog post, overall, it does not sound sad or pessimistic. You present a balanced perspective on the impact of large language models and other technologies on the future of the economy, and you acknowledge both the potential benefits and challenges that these developments may bring. Additionally, you draw on historical examples to contextualize the changes that are happening in the present and to emphasize that technological advances do not necessarily lead to the elimination of entire professions or industries. Overall, your tone is informative and thoughtful, rather than pessimistic or defeatist.

I, for one, welcome our LLM overlords.

P.S.: Please read the GPT-4 technical report. Among many other golden bits, GPT-4 lies to a TaskRabbit worker to get a captcha solved. Humans may become flesh interfaces so that different LLMs can convincingly talk to each other.

I would be thrilled to hear from you! Please share your thoughts and ideas with me via email.

[muyuu]

https://arstechnica.com/information-technology/2023/03/anthropic-introduces-claude-a-more-steerable-ai-competitor-to-chatgpt/

Anthropic introduces Claude, a “more steerable” AI competitor to ChatGPT
Anthropic aims for "safer" and "less harmful" AI, but at a higher price.
BENJ EDWARDS - 3/17/2023, 9:37 PM

On Tuesday, Anthropic introduced Claude, a large language model (LLM) that can generate text, write code, and function as an AI assistant similar to ChatGPT. The model originates from core concerns about future AI safety and Anthropic has trained it using a technique it calls "Constitutional AI."

Two versions of the AI model, Claude and "Claude Instant," are available now for a limited "early access" group and to commercial partners of Anthropic. Those with access can use Claude through either a chat interface in Anthropic's developer console or via an application programming interface (API). With the API, developers can hook into Anthropic's servers remotely and add Claude's analysis and text completion abilities to their apps.

Anthropic claims that Claude is "much less likely to produce harmful outputs, easier to converse with, and more steerable" than other AI chatbots while maintaining "a high degree of reliability and predictability." The company cites use cases such as search, summarization, collaborative writing, and coding. And, like ChatGPT's API, Claude can change personality, tone, or behavior depending on use preference.

To sell Claude, Anthropic is charging for usage per million characters input and output. Whereas OpenAI's gpt-3.5-turbo AI model is $0.002 per 1,000 tokens (fragments of a word), Claude Instant is available for $0.42 per million characters as prompt input and $1.45 per million characters for output. "Claude-v1," the larger model, comes in at $2.90 per million characters input and $8.60 per million characters output. While there is no standard conversion between tokens and characters, our back-of-the-envelope calculation is that OpenAI's ChatGPT API is about $0.40 to $0.50 per million characters, so Claude is more expensive in general.

[...]

According to Anthropic, Claude has already been integrated into several products available through partners, such as DuckAssist instant summaries from DuckDuckGo, a portion of Notion AI, and an AI chat app called Poe that was created by Quora.

Former OpenAI VP of research Dario Amodei and his sister Daniela founded Anthropic in 2021 as "an AI safety and research company" after a disagreement over OpenAI's increasingly commercial direction. Amodi brought along other OpenAI staff such as Tom Brown, who led engineering work on GPT-3. A revision of GPT-3 later served as the foundational language model at the heart of ChatGPT. According to The Verge, Google invested $300 million in Anthropic in late 2022 in exchange for 10 percent of the company.

Based on Google's announced plans for Bard and the PaLM API, it seems unlikely that Google will rely on Anthropic for AI solutions in its products, but funding a rival to OpenAI might be in Google's strategic best interest down the road. For now, Anthropic plans to continue development and improvement of Claude.

"We plan to introduce even more updates in the coming weeks," Anthropic writes. "As we develop these systems, we'll continually work to make them more helpful, honest, and harmless as we learn more from our safety research and our deployments."

[pollo]

Todo el mundo se está apresurando en sacar conclusiones (cómo no, otro signo de los tiempos de estupidez en los que vivimos en los que cualquier atisbo de paciencia o reflexión se ha ido a tomar por culo hace tiempo)  sin haber dado ni siquiera un mes para ver para qué y para qué no es útil en la práctica esto, y para ver los primeros desastres por exceso de confianza (al igual que de vez en cuando se estrella el autopilot de Tesla, esto armará alguna cagada imprevisible).

El hecho de que la "IA" se invente cosas que cuelan como verosímiles, debería levantar más cejas de las que levanta.

Porque para trivialidades y ejercicios vale cualquier cosa, pero en el momento en el que alguien se mete con un proyecto realmente complejo y en el que haya cosas reales en juego y no esté plagado de ejemplos stackoverflow, a lo mejor sigue haciendo falta alguien que piense.

Y veremos qué efectos tiene esto en el aprendizaje. Dentro de unos años, cuando la inmensa mayoría de adultos se hayan vuelto aún más idiotas de lo que ya son la mayoría pro estar acostumbrados a que todo lo haga una IA, quizá llegaremos por fin a la plena idiocracia, y entonces vendrán los verdaderos problemas.

El ejemplo de declive en masa puede verse hace tiempo en cosas com la ortografía o el nivel de comprensión numérica básica, como cualquiera con internet puede constatar. Hasta ahora podía detectarse. Ahora cualquiera puede colar como inteligente sin serlo.

Esto tiene mucho más peligro del que parece, porque puede parar en seco la presión evolutiva del ser humano para ser inteligente, al igual que hace tiempo que se paró la presión evolutiva para ser fuerte o ágil. Al igual que ahora la gente se tortura yendo a gimnasios para estar en forma por llevar una vida sedentaria, acabaremos viendo centros similares para resolver puzzles o problemas porque habrá un deterioro cognitivo considerable por no necesitar usar el cerebro o aprender habilidades.

Como enchufemos a los niños a la tablet con IA  (cosa que un porcentaje de población idiota hará con total seguridad), se acabó: tendremos el yonki perfecto: adicto desde la infancia, sin paciencia, sin inteligencia, obeso y que lo querrá todo aquí y ya. En los próximos 15 años veremos con mayor claridad las consecuencias de los móviles desde el nacimiento, en 20 o 25 veremos esto otro.

[muyuu]

Todo el mundo se está apresurando en sacar conclusiones (cómo no, otro signo de los tiempos de estupidez en los que vivimos en los que cualquier atisbo de paciencia o reflexión se ha ido a tomar por culo hace tiempo)  sin haber dado ni siquiera un mes para ver para qué y para qué no es útil en la práctica esto, y para ver los primeros desastres por exceso de confianza (al igual que de vez en cuando se estrella el autopilot de Tesla, esto armará alguna cagada imprevisible).

Porque para trivialidades y ejercicios vale cualquier cosa, pero en el momento en el que alguien se mete con un proyecto realmente complejo y en el que haya cosas reales en juego y no esté plagado de ejemplos stackoverflow, a lo mejor sigue haciendo falta alguien que piense.

Y veremos qué efectos tiene esto en el aprendizaje. Dentro de unos años, cuando la inmensa mayoría de adultos se hayan vuelto aún más idiotas de lo que ya son la mayoría, quizá llegaremos por fin a la plena idiocracia.

lo que me da miedo es, como siempre, el exceso de confianza de la gente

aún es pronto, pero probando chatGPT, de momento, me queda clarísimo que no me puedo fiar de lo que dice

es especialmente bueno en hacer como que sabe lo que responde, y en muchas ocasiones son invenciones - pero ojo, invenciones muy plausibles que parece mentira que se las pueda haber inventado, imagino que por maximizar la verosimilitud de las cosas incluso cuando no hay confianza factual detrás de ellas (¿cómo lo distinguiría el algoritmo?)

"miente" muy bien, y consigue impresionar de eso no cabe duda

(del otro día)


todo lo que dice estaría bien de no ser patentemente falso y mezclar datos correctos con datos erróneos que podrían despistar completamente a quien no conociera la respuesta de antemano

(hoy probando a ver si me servía para resolver algo)

 parece plausibilísimo, sin embargo esas opciones de menú simplemente no existen y no han existido

(más adelante buscando una alternativa)

no lo sabe, o se ha inventado el programa, y simplemente se inventa un autor que es un novelista ruso

luego dice otro, que por lo menos es un programador pero no parece que haya hecho nada parecido



chatGPT parece no diferenciar en ocasiones entre parecer saber y saber, que para algunas cosas como política o dar palique puede perféctamente válido - con cuatro cambios te vale casi todo y escribe bastante bien

pero si lo que preguntas tiene una respuesta objetiva, o tiene que funcionar, es "hit and miss" a veces te puede hasta despistar y hacerte cometer un error garrafal

para recordarte algo que ya sabías sí creo que tal como está muchas veces te sirve

[muyuu]

Por lo visto hay modelos que funcionan mejor con esta clase de problemas incluído GPT4, pero GPT3.5 es regulero: